Repository navigation
Add STS trace support (HuggingChat, llama.app) - #38
Conversation
Codex Review Summary: ✅ CompletedThis comment shows the latest Codex review activity on this pull request.
ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings. |
|
Looks good, thanks! |
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: e5c3c7cd79
ℹ️ About Codex in GitHub
Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".
| "role": "tool", | ||
| "tool_call_id": tool_call_id, | ||
| "name": tool_names_by_call_id.get(tool_call_id, "unknown_tool"), | ||
| "content": content, | ||
| } |
There was a problem hiding this comment.
Preserve failed STS tool results
When an STS role: "tool" event carries isError: true, this newly constructed message discards the flag and emits only the output text. The other trace converters normalize failed results to is_error: true, so these STS rows make failures indistinguishable from successful calls and can corrupt SFT supervision; copy the flag onto the normalized tool message.
Useful? React with 👍 / 👎.
Adds the Hugging Face Session Trace Simple Format, the STS JSONL that HuggingChat, llama.app and others export, so those traces can become SFT data. Before this they were detected as Pi and lost their tool calls.
Example traces: cfahlgren1/sts-format-traces